bad register name `%sil'

[Dave Cuthbert]

YONETANI Tomokazu wrote:
Is there anyone who knows how that thread in gcc-bugs list came to
a conclusion? At the end of the above log, someone named Richard Henderson
insists that the problem is in the sample code rather than in gcc, and that
using "=q" instead of "=r" solves the problem. It seems to me that that's
gcc team's decision, rather than fixing gcc by disabling the use of these
registers for IA32 platforms, not to mention that the same change has been
applied to atomic.h in FreeBSD-CURRENT months ago.
I don't follow gcc-bugs, but I think he was attempting to address the 
specific problem rather than the larger "gcc is choosing phantom 
registers" issue.

If you can narrow down the problem to a small sample, someone on the gcc 
team would probably be interested.

Dave